The way the software stack works, newer drivers work just fine with older CUDA versions. In fact a common usage mode is to install a CUDA version and use that for a couple of years, while periodically (say every three to six months) installing new driver packages to pick up bug fixes and performance improvements in the driver. So as far as I can tell, there is nothing wrong with your current configuration.
Each CUDA package comes packaged with a matching driver, and when you install CUDA this driver will be installed by default unless you tell the installer to keep your existing driver. Typically the driver packaged up with CUDA is the lowest driver version needed to support that particular CUDA version.
I have never used Tensorflow. I do not understand this error message, it is way too cryptic.
As I said, every CUDA version ships with a matching driver in the package, and that driver is installed by default. So if you perform a default installation of CUDA 10.2, you should have a driver that supports up to CUDA 10.2.
I am not sure how you wound up with CUDA 10.2 but a much more recent 11.1 driver. What I would try here is uninstall CUDA and re-install the full package (including the driver!) of whatever CUDA version Tensorflow desires.
